[ntp:questions] Re: Taming the pinball machine

David L. Mills mills at udel.edu
Tue Nov 18 16:38:11 UTC 2003


Roy,

While the problem with incorrect frequency estimation after a long
holdover period was fixed, the problem remains when sources clockhop a
considerable offset as sources come and go. The FLL design assumes the
principle source of error is clock frequency wander at very long poll
intervals, which is consistent with both theory and practice. It is not
clear in case of large time offset errors whether they are due to
ordinary thermal wander or some external cause. I still have no
explanation for the apparent huge frequency surge seen in previous
reports when in fact NTP was not adusting the clock at the time.

Dave

Roy wrote:
> 
> "David L. Mills" wrote in message news:<3FA9CC42.58299EA5 at udel.edu>...
> >
> > There have been several recent reports that certain configuration
> > scenarios experience somewhat naughty behavior when a valid
> > synchronization source is found after a holdover during which only the
> > local clock driver winds the clock. The result is a nasty frequency
> > surge that takes some time to dissipate.
> 
> FYI, I've seen a similar frequency surge without having the local
> clock driver configured.  Here are example loopstats generated with
> the original 4.2.0 release.  This looks like nominal adjustments
> between poll intervals 9 and 12:
> 
> 52946 32623.023 0.004414376 -122.306024 0.030704245 0.029020 9
> 52946 33647.020 0.003290657 -122.304455 0.029715764 0.025144 10
> 52946 34161.019 0.007188568 -121.970303 0.028514591 0.168489 10
> 52946 34672.028 0.011901096 -121.557778 0.025280663 0.252657 10
> 52946 35696.026 0.011828203 -121.500943 0.022135103 0.220645 10
> 52946 36721.021 0.012520656 -121.380475 0.019331504 0.200353 11
> 52946 37744.019 0.014827820 -121.123208 0.016837202 0.215992 11
> 52946 38767.020 0.011270369 -121.417935 0.014913665 0.238129 11
> 52946 40815.018 0.012852483 -121.201459 0.013160134 0.232905 12
> 
> Between 14 and 16, the frequency adjustments became more pronounced:
> 
> 52947 3570.023 0.027249104 -120.865829 0.039217946 0.568073 14
> 52947 19955.025 0.056365251 -119.926354 0.041902504 0.680208 15
> 52947 36339.022 0.005975020 -120.916182 0.050293743 0.769384 15
> 52947 52725.027 -0.052044137 -122.093038 0.054006568 0.888938 15
> 52947 85492.026 0.004573343 -121.005418 0.057051619 0.942543 16
> 52948 31862.023 0.062506310 -120.119140 0.058408479 0.928796 15
> 52948 64632.024 -0.048148949 -121.206550 0.076392624 0.970882 15
> 52949 10999.026 -0.155267736 -122.326981 0.086870017 1.010347 15
> 52949 43767.024 -0.099776332 -121.858918 0.082285588 0.905744 16
> 
> By the time the poll interval climbs to 17, the frequencies and time
> offsets are jumping all over the place.
> 
> 52949 76535.023 -0.042154193 -121.026594 0.078694650 0.887958 16
> 52950 22905.022 0.013110888 -120.203082 0.076081072 0.872293 17
> 52951 67581.024 -0.419374902 -124.594881 0.128691301 2.322207 16
> 52954 70497.039 1.383547201 -120.239999 0.446881767 2.964074 16
> 52954 70514.031 1.371672806 -124.190652 0.580923284 3.239015 17
> 52956 28768.040 1.469172093 -122.809917 0.528822809 2.888775 17
> 52956 28774.108 1.524055147 -86.218412 0.479233267 18.466005 17
> 52957 73448.113 1.612414477 34.247289 0.429699568 62.319670 16
> 
> This was the last loopstat generated, since I stopped ntpd shortly
> after this.  I grew tired of watching the pinball machine.  When ntpd
> was stopped, the time was nearly 26 seconds off.  I've now restarted
> again with the revised algorithms.
> 
> > 2. Really DO use burst mode (both iburst and burst keywords) if you
> > intend to operate much above the default maximum.
> 
> When burst mode is used, that hits the servers with 8 packets each
> poll rather than just one.  That doesn't seem to help reduce the
> server load much. :{
> Hmmm, I wonder which configuration gives better results:
>     maxpoll 17 with burst
>     maxpoll 14 without burst
> 
> Have a great time,
> 
> 
> roy
> --
> The suespammers.org mail server is located in California.  Please do
> not send unsolicited bulk e-mail or unsolicited commercial e-mail to
> my suespammers.org address or any of my other addresses.  These are my
> opinions, not necessarily my employer's.



More information about the questions mailing list